JP5371676B2 - RECOMMENDED CONTENT EXTRACTION DEVICE, RECOMMENDED CONTENT EXTRACTION METHOD, AND RECOMMENDED CONTENT EXTRACTION PROGRAM - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To effectively extract a recommended content in a specific service having no use histories of a user. <P>SOLUTION: Use history information having user identification information identifying a plurality of user terminals connected via a network, associated with content identification information of contents transmitted to the user terminals corresponding to the user identification information, is stored for each of service identification information for identifying a plurality of service providers connected via the network, and similarities of the use history information between users are calculated per piece of the service identification information, and an integrated similarity being an average value of the plurality of similarities is calculated, and users are clustered on the basis of the calculated integrated similarity, and a content of recommended object service included in use history information of users classified to the same cluster as a recommendation object user is determined to be content identification information to be recommended to the recommendation object user. <P>COPYRIGHT: (C)2011,JPO&amp;INPIT

Description

  The present invention relates to a technique for extracting content recommended to a user.

In recent years, a service for transmitting various information to a user's computer terminal via a network such as the Internet has been provided. For example, various services such as a shopping service, a video distribution service, a news distribution service, an auction service, and a web mail service can be used via a network. Here, for example, in a shopping service that sells content (products), other recommended content may be determined according to the content usage history for each user, and information on the recommended content may be transmitted. For example, a correlation between a user's usage history and a usage history of another user in a specific service is analyzed, and content used by another user having a high correlation with the user is recommended to the specific user. According to such content recommendation, the user can obtain useful information, and the service provider can effectively advertise the content by his service.
Japanese Patent Application Laid-Open No. 2004-228561 describes a technique for calculating a user's similarity based on the correlation of users having usage histories in a plurality of domains having different data formats, and making a recommendation based on the calculated similarity.

Special table 2008-502028 gazette

  However, in order to determine the content recommended to the user in the conventional technology, it is necessary to accumulate the user's usage history in that service. That is, for users who have accumulated usage histories for specific services, it is possible to determine recommended content based on correlation with other users with similar usage histories, but usage histories are accumulated. Since no other users with high correlation of usage histories cannot be determined for users who do not, the recommended content cannot be determined. Even in such a case, it is desirable to determine effective content recommended to the user. For example, when a service provider that provides a specific service for which usage history has been accumulated has started providing another new service, in order to encourage the use of the new service, the user who has no usage history in the new service It is desirable to recommend appropriate content.

  The present invention has been made in view of such a situation, and a recommended content extraction device, a recommended content extraction method, and a method for extracting recommended content that are effective for a specific service without a user's usage history. The present invention relates to a recommended content extraction program.

In order to solve the above-described problem, the present invention provides a plurality of service providing apparatuses that transmit a content requested from a user terminal among a plurality of determined contents to a plurality of user terminals connected via a network. Connected recommended content extraction device, for each service identification information for identifying a service providing device, user identification information for identifying a user of the user terminal, and content for identifying content transmitted to the user terminal by the service providing device A usage history storage unit that stores usage history information associated with identification information, and a cluster identification that identifies a user cluster determined by clustering based on the usage history information corresponding to the user identification information for each user identification information A clustering information storage unit for storing information in association with each other; For each identification information, based on a predetermined distance measure, a similarity calculation unit that calculates the similarity of the usage history information between the user and the cluster centroid, and a plurality of items calculated for each service identification information by the similarity calculation unit An integrated similarity calculating unit that calculates an integrated similarity that is an average value of the similarities, and clustering for each user based on the integrated similarity calculated by the integrated similarity calculating unit to determine a cluster corresponding to the user A clustering unit that associates the user identification information for identifying the user with the cluster identification information for identifying the cluster corresponding to the user identification information and stores them in the clustering information storage unit, and recommends content among the plurality of user terminals. The recommendation target user identification information to which the recommendation target user identification information for identifying the user terminal of the recommendation target user that is the target to be input is input. A recommendation target service identification information input unit for inputting recommendation target service identification information for identifying a service provision device that transmits content recommended to a user terminal of a recommendation target user among a plurality of service providing devices; The cluster identification information associated with the recommendation target user identification information input to the recommendation target user identification information input unit is read from the clustering information storage unit, and among the user identification information associated with the read cluster identification information, the recommendation target Content identification that reads user identification information different from the user identification information and recommends content identification information stored in the usage history storage unit in association with the read user identification information and recommendation target service identification information to the recommendation target user A recommended content extraction unit that determines that the information is information. It is a sign.

  Further, according to the present invention, the integrated similarity calculation unit described above performs weighting by multiplying a plurality of similarities calculated for each service identification information by the similarity calculation unit by a weighting factor predetermined for each service identification information. An integrated similarity that is an average value of similarities multiplied by a coefficient is calculated.

  Further, according to the present invention, the similarity calculation unit described above is based on any one of a plurality of distance measures including a Euclidean distance and a cosine similarity predetermined for each service identification information. The degree of similarity of usage history information is calculated.

In addition, the present invention provides, for each service identification information for identifying a plurality of service providing apparatuses that transmit a requested content from a plurality of user terminals connected via a network among a plurality of determined contents. A usage history storage unit that stores usage history information in which user identification information for identifying a user and content identification information for identifying content transmitted to a user terminal by a service providing device are associated, and for each user identification information And a clustering information storage unit for storing cluster identification information for identifying user clusters determined by clustering based on the use history information corresponding to the user identification information, and storing the associated information. A similarity calculation unit for each service identification information; A step of calculating the similarity of the usage history information between the user and the cluster centroid based on the determined distance measure, and a plurality of similarities calculated by the integrated similarity calculating unit for each service identification information by the similarity calculating unit A step of calculating an integrated similarity that is an average value of degrees, and a clustering unit performs clustering for each user based on the integrated similarity calculated by the integrated similarity calculating unit to determine a cluster corresponding to the user, A step of storing user identification information for identifying a user and cluster identification information for identifying a cluster corresponding to the user identification information in a clustering information storage unit, and a plurality of user terminals in the recommendation target user identification information input unit Among the recommendation target user identifications for identifying the user terminals of the recommendation target users who are content recommendation targets. And a recommendation target service identification information identifying a service provision device that transmits content to be recommended to the user terminal of the recommendation target user among the plurality of service provision devices in the step of inputting information and the recommendation target service identification information input unit. The input step and the recommended content extraction unit read cluster identification information associated with the recommendation target user identification information input to the recommendation target user identification information input unit from the clustering information storage unit, and read the cluster identification information Content that is stored in the usage history storage unit in association with the read user identification information and recommendation target service identification information is read out of the associated user identification information that is different from the recommendation target user identification information Content identification information that recommends the identification information to the recommendation target user And a step of determining.

In addition, the present invention provides, for each service identification information for identifying a plurality of service providing apparatuses that transmit a requested content from a plurality of user terminals connected via a network among a plurality of determined contents. A usage history storage unit that stores usage history information in which user identification information for identifying a user and content identification information for identifying content transmitted to a user terminal by a service providing device are associated, and for each user identification information A clustering information storage unit that stores cluster identification information that identifies user clusters determined by clustering based on clustering based on usage history information corresponding to user identification information, and is stored in a computer of a recommended content extraction device, The similarity calculation unit determines a predetermined distance for each service identification information. A step of calculating the similarity of the usage history information between the user and the cluster centroid based on the degree, and the integrated similarity calculating unit calculates an average value of the plurality of similarities calculated for each service identification information by the similarity calculating unit A step of calculating the integrated similarity and a clustering unit that performs clustering for each user based on the integrated similarity calculated by the integrated similarity calculating unit, determines a cluster corresponding to the user, and identifies the user The user identification information and the cluster identification information for identifying the cluster corresponding to the user identification information are stored in the clustering information storage unit in association with each other, and the recommendation target user identification information input unit includes the content among the plurality of user terminals. Recommendation target user identification information for identifying the user terminal of the recommendation target user who is the target of recommending And a recommendation target service identification information input unit that receives recommendation target service identification information for identifying a service provision device that transmits content recommended to a user terminal of a recommendation target user among a plurality of service provision devices. The recommended content extraction unit reads the cluster identification information associated with the recommendation target user identification information input to the recommendation target user identification information input unit from the clustering information storage unit, and the user associated with the read cluster identification information Among the identification information, the user identification information different from the recommendation target user identification information is read, and the content identification information stored in the usage history storage unit in association with the read user identification information and the recommendation target service identification information is recommended. Step that determines content identification information recommended for the target user This is a recommended content extraction program that executes the program.

  As described above, according to the present invention, for each service identification information for identifying a plurality of service providing apparatuses connected via a network, user identification information for identifying a plurality of user terminals connected via a network And the usage history information associated with the content identification information of the content transmitted to the user terminal corresponding to the user identification information, and the similarity of the usage history information between the user and the cluster centroid for each service identification information Calculating the degree of integration, calculating an integrated similarity that is an average value of a plurality of the similarities, performing clustering for each user based on the calculated integrated similarity, and classifying the same cluster as the recommended user It is content identification information that recommends the content of the recommendation target service included in the user usage history information to the recommendation target user. Based on the usage history of a plurality of different services, it is possible to extract recommended content for a specific service among a plurality of services, and effective recommendation for users who have no usage history in a specific service Content can be extracted.

It is a block diagram which shows the structural example of the service provision system by one Embodiment of this invention. It is a figure which shows the example of the collection of the users who utilize the service by one Embodiment of this invention. It is a figure which shows the example of data of the usage history information memorize | stored in the usage history memory | storage part by one Embodiment of this invention. It is a figure which shows the example of data of the user usage history matrix memorize | stored in the user usage history matrix memory | storage part by one Embodiment of this invention. It is a figure which shows the example of data of the user and gravity center similarity matrix memorize | stored in the user and gravity center similarity matrix memory | storage part by one Embodiment of this invention. It is a figure which shows the example of data of the clustering result memorize | stored in the clustering information storage part by one Embodiment of this invention. It is a figure which shows the example of the recommended content extracted by one Embodiment of this invention. 6 is a flowchart illustrating an operation example in which a service providing system according to an exemplary embodiment of the present invention determines recommended content. It is a flowchart which shows the specific operation example of the utilization log | history storage process by one Embodiment of this invention. It is a flowchart which shows the specific operation example of the user history matrix creation process by one Embodiment of this invention. It is a flowchart which shows the specific operation example of the clustering process by one Embodiment of this invention. It is a flowchart which shows the specific operation example of the recommendation content extraction process by one Embodiment of this invention.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
FIG. 1 is a block diagram illustrating a configuration of a service providing system 1 according to the present embodiment. The service providing system 1 uses other service providing apparatuses 200-j (j is different from i) based on the usage history of the content provided to the user terminal 100 by the service providing apparatus 200-i (i is an arbitrary number). The content of the service provided by (arbitrary number) is recommended to the user terminal 100-k (k is an arbitrary number). For example, FIG. 2 is a diagram illustrating an example of a set of users who use different services. The service provided by the service providing apparatus 200-1 is referred to as service 1 (movie distribution service), and the service provided by the service providing apparatus 200-2 is referred to as service 2 (news article distribution service). Here, the service 1 is used by many users, and usage histories of a plurality of users are accumulated. On the other hand, it is assumed that the service 2 is a newly launched service, has a smaller number of users than the service 1, and also stores less usage history.

  In such a case, the service providing system 1 according to the present embodiment recommends content suitable for the user from the content provided by the service 2 to the user who uses only the service 1. Thus, useful information can be provided to a user who uses only the service 1, and effective advertisement for the service provider of the service 2 can be performed. In addition, a user who uses only the service 1 can provide an opportunity to use the service 2.

  Returning to FIG. 1, the service providing system 1 includes a plurality of user terminals 100 (user terminals 100-1, user terminals 100-2,...) And a plurality of service providing apparatuses 200 (service providing apparatuses 200-1, services). Providing device 200-2, service providing device 200-3,...) And recommended content extracting device 300. Here, since the plurality of user terminals 100 have the same configuration, they are described as the user terminals 100 when it is not necessary to distinguish between them. Similarly, since the plurality of service providing apparatuses 200 have the same configuration, the service providing apparatus 200 will be described as a service providing apparatus 200 when there is no need to distinguish between them.

  The user terminal 100 is a computer device connected to a plurality of service providing devices 200 via a network. The user terminal 100 has an Internet browser function, for example, and inputs a user ID or password registered in advance in the service providing apparatus 200 to log in to a service provided by the service providing apparatus 200. Further, the user terminal 100 transmits a content information acquisition request to the service providing apparatus 200 in response to an input of operation information from the user, and receives content information transmitted in response from the service providing apparatus 200.

  The service providing apparatus 200 is a computer apparatus that provides a service for transmitting content requested from the user terminal 100 among a plurality of determined contents. The service providing device 200 includes, for example, services such as shopping services, video distribution services, news distribution services, auction services, web mail services, map information services, online dictionary services, links to these services, search engines, and the like. Provide provided portal site services. In the present embodiment, the service providing apparatus 200-1 provides a movie distribution service that transmits a movie (movie content) in response to a movie acquisition request transmitted from the user terminal 100, and the service providing apparatus 200-2 includes: A news article distribution service that transmits a news article in response to a news article acquisition request transmitted from the user terminal 100 is provided. The service providing apparatus 200 includes a content storage unit 210 and a service providing unit 220.

The content storage unit 210 stores a plurality of predetermined contents. For example, the content storage unit 210-1 of the service providing apparatus 200-1 that provides a movie distribution service stores moving image contents of a plurality of movies. A plurality of news articles are stored in the content storage unit 210-2 of the service providing apparatus 200-2 that provides the news article distribution service.
In response to a content acquisition request transmitted from the user terminal 100 via the network, the service providing unit 220 reads the content stored in the content storage unit 210 and sends it to the user terminal 100 that is the transmission source of the content acquisition request. Send.

  The recommended content extraction device 300 is a computer device that extracts recommended content according to a user in a specific service provided by a plurality of service providing devices 200, and includes a storage unit 310, a usage history acquisition unit 320, and a user usage history. The matrix generation unit 330, the similarity calculation unit 340, the integrated similarity calculation unit 350, the clustering unit 360, the recommendation target user identification information input unit 370, the recommendation target service identification information input unit 380, and the recommended content extraction unit 390.

  The storage unit 310 includes a plurality of usage history storage units 311 (a usage history storage unit 311-1, a usage history storage unit 311-2, a usage history storage unit 311-3,...) And a user usage history matrix storage unit 312. And a user / centroid similarity matrix storage unit 313 and a clustering information storage unit 314, which store information for extracting recommended content according to the user. Here, the usage history information corresponding to each service is stored in the plurality of usage history storage units 311. However, since usage history information in the same data format is stored, it is not particularly necessary to distinguish between them. Will be described as the usage history storage unit 311.

  In the plurality of usage history storage units 311, for each service ID for identifying the corresponding service providing apparatus 200, the user ID for identifying the user terminal 100 and the content transmitted to the user terminal 100 by the service providing apparatus 200 are stored. Usage history information associated with a content ID to be identified is stored. FIG. 3 is a diagram illustrating a data example of usage history information stored in the usage history storage unit 311. For example, the symbol a indicates a data example of usage history information stored in the usage history storage unit 311-1 corresponding to the service 1, and the symbol b is stored in the usage history storage unit 311-2 corresponding to the service 2. An example of data of usage history information is shown, and symbol n is a data example of usage history information stored in the usage history storage unit 311 -N corresponding to service N (N is a number corresponding to the number of service providing apparatuses 200). Indicates.

  Here, each user ID memorize | stored in the utilization log | history memory | storage part 311 for every service is a common user ID which identifies the same user terminal 100. FIG. User IDs that are common between such different services may be, for example, user IDs registered in portal sites that are common to both services, and if the service providers are the same, any service ID The user ID registered in can be used for other services. Or you may make it identify the identity of the user between different services by user ID like OpenID which can be used in common across sites. The content ID is identification information for identifying different content for each service.

  The user usage history matrix storage unit 312 stores, for each service ID, information including a user ID and information indicating whether the content provided by the corresponding service ID is provided. FIG. 4 is a diagram illustrating a data example of the user usage history matrix stored in the user usage history matrix storage unit 312. As described above, the user usage history matrix storage unit 312 in the present embodiment includes Boolean information indicating whether or not the user has used the content in association with the content ID and the user ID for each service ID. A user usage history matrix is stored. Here, an example is shown in which Boolean data having a row as a user ID and a column as a content ID for each service is stored. The data in the user usage history matrix storage unit 312 includes, for example, an amount of content used ( For example, the number of times the user referred to the news article, the time), and the like may be stored.

  The user / centroid similarity matrix storage unit 313 stores, for each user ID, a user / centroid similarity matrix in which information indicating similarity to the cluster centroid is associated. FIG. 5 is a diagram illustrating a data example of the user / centroid similarity matrix stored in the user / centroid similarity matrix storage unit 313. Here, the user / centroid similarity matrix storage unit 313 integrates the user / centroid similarity matrix calculated for each service indicated by symbols a and b and the similarity of all services indicated by symbol z. A similarity user / centroid similarity matrix is stored. In this example, an example of data when cosine similarity is applied as a distance scale is shown. In this case, the degree of similarity of the user is represented by a value from 0 to 1, and the higher the value, the higher the degree of similarity.

  The clustering information storage unit 314 stores, for each user ID, a cluster ID that identifies a user cluster determined by clustering based on usage history information corresponding to the user ID. FIG. 6 is a diagram illustrating a data example of the clustering result stored in the clustering information storage unit 314. Here, cluster ID “1” is associated with user ID “00001”, cluster ID “2” is associated with user ID “00002”, and user ID “00003” is associated with. An example in which the cluster ID “1” is associated is shown.

  When the user terminal 100 transmits an information acquisition request to the service providing apparatus 200 and the service providing apparatus 200 transmits content in response to the acquisition request, the usage history acquisition unit 320 transmits the acquisition request. A user ID corresponding to the service ID, a service ID corresponding to the service providing apparatus 200 that transmitted the content, and a content ID corresponding to the content transmitted by the service providing apparatus 200, and a usage history storage unit corresponding to the service ID 311 is stored.

  The user usage history matrix generation unit 330 reads the usage history information stored for each user from the usage history storage unit 311, generates the above-described user usage history matrix in which the usage history information for each service is stored, and the user usage Store in the history matrix storage unit 312.

  Based on the user usage history matrix stored in the user usage history matrix storage unit 312 by the user usage history matrix generation unit 330, the similarity calculation unit 340 is connected to the user based on a predetermined distance measure for each service ID. The similarity of the usage history information between the cluster centroids is calculated. Here, the distance scale is, for example, a Euclidean distance predetermined for each service ID according to the nature of the usage history information, such as whether the information included in the usage history information is a discrete value or a continuous value. Any one of a plurality of distance measures such as cosine similarity can be applied. For example, as in this embodiment, in the user usage history matrix storage unit 312, a Boolean type indicating whether or not the user has used the content in association with the content ID and the user ID for each service ID as described above. In the case where the information is stored, it is desirable to apply the cosine similarity that handles binary values. On the other hand, the user usage history matrix storage unit 312 stores, for example, the amount of content used (for example, the number of times the user has referred to a news article, the time), or the like, or the number of content is large and the number of dimensions of the space When there are many, Euclidean distance may be applied.

  For example, when the cosine similarity is applied as a distance measure, the similarity calculation unit 340 sets the usage history information of the user in the service as a vector having N elements when the number of contents of the service to be processed is N. Then, an angle (cosine) between the vector based on the user's usage history information and the center of gravity of each cluster is calculated, and the calculated angle is set as the similarity to each cluster of the user in the service to be processed. The similarity calculation unit 340 stores the similarity of the usage history information calculated for each service in the user / centroid similarity matrix storage unit 313 as a user / centroid similarity matrix for each service.

  Based on the user / centroid similarity matrix calculated for each service ID by the similarity calculation unit 340, the integrated similarity calculation unit 350 is an integration that is an average value of similarities between each user and each cluster for each service ID. Calculate similarity. Here, when the average value is calculated, if there is a user / centroid similarity matrix of a service with no user usage history (for example, the user with the user ID “00002” indicated by the symbol b in FIG. 5), the service Does not take into account the average value. For example, the integrated similarity calculation unit 350 adds the similarity values in the user / centroid similarity matrix of a service for which there is a user usage history, and the user / centroid similarity matrix of the service for which no user usage history exists. The addition of the similarity is skipped, and the average value is calculated by dividing by the number of user / centroid similarity matrices of the services for which the user usage history exists. By calculating the integrated similarity in this way, it is possible to prevent the use history of a service that has no use history, that is, a service that is not used, from affecting the clustering. The integrated similarity calculation unit 350 stores the user / centroid similarity matrix of the calculated integrated similarity in the user / centroid similarity matrix storage unit 313.

  Here, the integrated similarity calculation unit 350 multiplies a plurality of similarities calculated for each service ID by the similarity calculation unit 340 by a weighting coefficient predetermined for each service ID, and multiplies the weighting coefficient by the similarity. You may make it calculate the integrated similarity which is an average value of degree. The weight continuation coefficient predetermined for each service ID is stored in advance in the storage area of the integrated similarity calculation unit 350. For example, by setting a value of 1 or less to the weighting coefficient of a service that has just started up, has not passed days since the service started, and has little accumulation of usage history information, the user's preference is not fully expressed. It is possible to reduce the influence of the usage history information considered to be the integrated similarity, and to calculate the integrated similarity with high reliability. On the other hand, by setting a value of 1 or more to the weighting coefficient of the service in which a sufficient amount of usage history information is stored, the usage history information that is considered to sufficiently express the user's preference is displayed in the integrated similarity. It is possible to calculate the integrated similarity with high reliability by increasing the influence. In addition, the integrated similarity calculation unit 350, when the similarity calculation unit 340 calculates the similarity based on a different distance scale for each service, determines the similarity between predetermined similarity stored in its own storage area. The integrated similarity may be calculated by multiplying the correction coefficient.

  Based on the integrated similarity user / centroid similarity matrix stored in the user / centroid similarity matrix storage unit 313, the clustering unit 360 performs clustering (cluster analysis) for each user to determine a cluster corresponding to the user. Then, the user ID for identifying the user and the cluster ID for identifying the cluster corresponding to the user ID are associated with each other and stored in the clustering information storage unit 314. The clustering unit 360 classifies users by a technique such as hierarchical clustering such as the Ward method that classifies users hierarchically, or non-hierarchical clustering such as the K-average method that classifies users into a predetermined number of clusters. To do. In the present embodiment, an example in which clustering is performed by the K-average method will be described.

  In the K-average method, for example, the clustering unit 360 randomly extracts user IDs of a predetermined number of clusters, and uses usage history information corresponding to the extracted user IDs as cluster centroids. Then, the similarity calculation unit 340 and the integrated similarity calculation unit 350 calculate the integrated similarity with each cluster centroid for each user. The clustering unit 360 assigns the cluster ID having the highest integrated similarity to the user. Then, the clustering unit 360 updates the cluster centroid and performs similarity calculation and cluster ID assignment processing until there is no change in the cluster centroid.

  The recommendation target user identification information input unit 370 receives a recommendation target user ID for identifying the user terminal 100 of the recommendation target user that is a target for recommending the content among the plurality of user terminals 100. For example, when the information acquisition request is transmitted from the user terminal 100 to the service providing apparatus 200, the recommendation target user identification information input unit 370 sets the user ID corresponding to the user terminal 100 that transmitted the acquisition request to the service providing apparatus 200. The acquired user ID may be input as the recommendation target user ID. Alternatively, the recommendation target user identification information input unit 370 includes an input device such as a keyboard and a mouse so that a recommendation target user ID for identifying a specific user is input from an administrator of the recommended content extraction apparatus 300 or the like. Anyway.

  The recommendation target service identification information input unit 380 receives a recommendation target service ID for identifying the service providing apparatus 200 that transmits the recommended content to the user terminal 100 of the recommendation target user among the plurality of service providing apparatuses 200. For example, when the service providing apparatus 200 receives an information acquisition request from the user terminal 100, the recommendation target service identification information input unit 380 acquires and acquires a service ID corresponding to the service providing apparatus 200 that has received the acquisition request. The service ID may be input as the recommendation target service ID. Alternatively, the recommendation target service identification information input unit 380 includes an input device such as a keyboard and a mouse so that a recommendation target service ID for identifying a specific service is input from an administrator of the recommended content extraction apparatus 300 or the like. Anyway.

  The recommended content extraction unit 390 reads the cluster ID associated with the recommendation target user ID input to the recommendation target service identification information input unit 380 from the clustering information storage unit 314, and the user ID associated with the read cluster ID. Among them, content that recommends a user ID that is different from the recommendation target user ID and recommends the content ID stored in the usage history storage unit 311 in association with the read user ID and the recommendation target service ID to the recommendation target user. It is determined that it is an ID. Here, for example, in the recommendation target service ID, the recommended content extraction unit 390 determines that the content ID most purchased by other users in the same cluster is the recommended content ID. Alternatively, for example, the most recently sold content ID may be determined as the recommended content ID.

  FIG. 7 is a diagram illustrating an example of recommended content extracted by the recommended content extraction unit 390. For example, the usage history information of the service 1 stored in the usage history storage unit 311-1 indicated by the symbol a and the usage history information of the service 2 stored in the usage history storage unit 311-2 indicated by the symbol b Based on this, the similarity calculation unit 340 and the integrated similarity calculation unit 350 calculate the integrated similarity, and based on the calculated integrated similarity, the clustering unit 360 performs clustering as indicated by reference numeral b1 (b2). To do. Here, user 1 and user 2 are clustered in cluster 1, user 3, user 4 and user 5 are clustered in cluster 2, and user 6, user 7 and user 8 are clustered in cluster 3. For user 5, service 1 has a usage history, but service 2 has no usage history. Therefore, the recommended content extraction unit 390, for the user 5, based on the usage history information in the service 2 of the user 3 and the user 4 that are in the same cluster, the content 3 purchased by the user 3 and the user 4 The content 4 is determined to be a recommended content ID. The recommended content extraction unit 390 transmits the determined recommended content ID to the user terminal 100 corresponding to the user 5.

  Next, an operation example of the service providing system 1 according to the present embodiment will be described. FIG. 8 is a flowchart illustrating an operation example in which the service providing system 1 determines recommended content. First, the usage history acquisition unit 320 stores user usage histories in a plurality of services in the usage history storage unit 311 (step S100). FIG. 9 is a flowchart illustrating a specific operation example of the usage history storage process. For example, when the user terminal 100-1 transmits a content acquisition request to the service providing apparatus 200-1, the service providing apparatus 200-1 transmits the content corresponding to the acquisition request transmitted from the user terminal 100-1 to the user terminal. 100-1 is transmitted (step S101).

  When the usage history acquisition unit 320 of the recommended content extraction device 300 detects that the content is transmitted from the service providing device 200-1 to the user terminal 100-1, the user ID corresponding to the user terminal 100-1 and the service provision are provided. A service ID corresponding to the device 200-1 and a content ID corresponding to the content transmitted by the service providing device 200-1 to the user terminal 100-1 are acquired (step S102). The usage history acquisition unit 320 stores the service ID, the user ID, and the content ID in association with each other in the usage history storage unit 311-1 that is the usage history storage unit 311 corresponding to the acquired service ID (step S103). ). The usage history acquisition unit 320 performs such usage history storage processing every time content is transmitted from the service providing apparatus 200 to the user terminal 100.

  Returning to FIG. 8, the user usage history matrix generation unit 330 creates a user history matrix based on the usage history information stored in the usage history storage unit 311 (step S <b> 200). FIG. 10 is a flowchart illustrating a specific operation example of the user history matrix creation process. The user usage history matrix generation unit 330 reads usage history information corresponding to a specific user ID from the plurality of usage history storage units 311 (step S201). The user usage history matrix generation unit 330 determines whether or not a user ID corresponding to the read usage history information exists in the user history matrix stored in the user usage history matrix storage unit 312 (step S202). When it is determined that the user ID corresponding to the read usage history information exists in the user history matrix (step S202: YES), the user usage history matrix generation unit 330 stores the read usage history information in the user usage history matrix storage unit 312. The data is overwritten and stored (step S203). If it is determined that the user ID corresponding to the read usage history information does not exist in the user history matrix (step S202: NO), a row corresponding to the user ID is created in the user usage history matrix storage unit 312 and the read usage history is stored. The information is stored in the user usage history matrix storage unit 312 (step S204).

  Returning to FIG. 8, the similarity calculation unit 340, the integrated similarity calculation unit 350, and the clustering unit 360 are based on the user history matrix stored in the user usage history matrix storage unit 312 and each user and each cluster centroid. The similarity is calculated and clustering processing is performed (step S300). FIG. 11 is a flowchart illustrating a specific operation example of the clustering process. The clustering unit 360 initializes the cluster centroid (step S301). Here, for example, the clustering unit 360 randomly extracts a predetermined number (for example, 10) of user IDs from among a plurality of user IDs stored in the user usage history matrix storage unit 312, and extracts the extracted user IDs. Is stored in the user usage history matrix storage unit 312 as the cluster centroid. Here, for example, the clustering unit 360 adds a data row that becomes a cluster centroid to the user usage history matrix stored in the user usage history matrix storage unit 312 and stores the usage history information that is the cluster centroid.

  Then, the similarity calculation unit 340 and the integrated similarity calculation unit 350 calculate the similarity between each user and each cluster centroid (or the user when the use history information of a specific user is applied as the cluster centroid). (Step S302). Specifically, the similarity calculation unit 340 reads user usage history information stored in the user usage history matrix storage unit 312 (step S302). The similarity calculation unit 340 calculates the similarity between each user stored in the user usage history matrix generation unit 330 and the usage history information corresponding to the data row that is the cluster centroid, and generates a user / centroid similarity matrix Then, it is stored in the user / centroid similarity matrix storage unit 313 (step S304). The integrated similarity calculation unit 350 calculates the integrated similarity by integrating the user / centroid similarity matrix for each service calculated by the similarity calculation unit 340 and stored in the user / centroid similarity matrix storage unit 313. The center-of-gravity similarity matrix storage unit 313 stores an integrated similarity user-centroid similarity matrix (step S305).

  The clustering unit 360 determines the nearest cluster centroid for each user based on the user / centroid similarity matrix of the integrated similarity stored in the user / centroid similarity matrix storage unit 313, and corresponds to each user ID. Clustering processing for determining a cluster is performed (step S306). The clustering unit 360 calculates a cluster centroid for each service based on the determined cluster for each user ID and usage history information for each user ID, and stores the cluster centroid in the user / centroid similarity matrix storage unit 313 (step S307). . Here, the clustering unit 360 determines whether any of the calculated plurality of cluster centroids has changed compared to the previously calculated cluster centroid (step S308).

  If the clustering unit 360 determines that the cluster centroid has changed (step S308: YES), the process returns to step S302. If it is determined that there is no change in the cluster center of gravity (step S308: NO), the cluster ID for each user is stored in association with the clustering information storage unit 314 (step S309).

  Returning to FIG. 8, the recommended content extraction unit 390 determines a recommended content ID for the recommended target user ID based on the recommended target user ID and the recommended target service ID (step S400). FIG. 12 is a flowchart illustrating a specific operation example of the recommended content extraction process. The recommendation target user identification information input unit 370 receives an input of a recommendation target user ID, and the recommendation target service identification information input unit 380 receives an input of a recommendation target service ID (step S401).

  The recommended content extraction unit 390 extracts other user IDs belonging to the cluster corresponding to the recommendation target user from the clustering information storage unit 314 (step S402). The recommended content extraction unit 390 reads the usage history information associated with the extracted user ID from the usage history storage unit 311 corresponding to the recommended service, and among the read usage history information, the usage frequency (number of purchases) A predetermined number (for example, 10) of content IDs with a large number is extracted (step S403). The recommended content extraction unit 390 transmits the extracted recommended content ID to the user terminal 100 corresponding to the recommendation target user (step S404). Here, for example, the recommended content extraction unit 390 reads information such as a name corresponding to the recommended content ID from the service providing apparatus 200 corresponding to the recommendation target service ID, and transmits the information to the user terminal 100. The user terminal 100 displays information corresponding to the recommended content ID. In this way, for example, another user classified into the same cluster as a user who likes movies of the baseball genre in the movie distribution service (service 1), and the usage history in the news article distribution service (service 2) It is possible to recommend the content of news articles about baseball in the service 2 to users who do not have any.

  In this embodiment, movie content and news article content have been described as examples of content stored in the content storage unit 210 and recommended by the recommended content extraction unit 390. It can be content. For example, it is possible to recommend, as content, information corresponding to a recommended product corresponding to a product purchased by a user as content, with information on a product sold in a shopping service for mail order sales of the product.

  In the present embodiment, as the cluster centroid initialization process, the clustering unit 360 randomly extracts a predetermined number of user IDs, and stores usage calendar information corresponding to the extracted user IDs in the user usage history matrix. However, for example, users are randomly classified into a predetermined number of clusters, and an average value of user usage history belonging to the same cluster is stored in the user usage history matrix as a cluster centroid. Also good.

  As described above, according to the recommended content extraction apparatus 300 of the present embodiment, not only the extraction of recommended content in a service in which a user's usage history is accumulated, but also an effect in a specific service without a user's usage history. It is possible to extract recommended content. Thus, for example, a service provider company that provides a plurality of services grasps user preferences based on usage histories of services already used by many users, and content groups in other services that do not have usage histories It is possible to recommend content suitable for the user from among the above. That is, among a plurality of services whose users can be identified, users acquired in a certain service can be guided to other services, and an increase in content usage by the user can be expected. In addition, the use history of content in services that have no or little usage history can be obtained, so that more accurate clustering can be performed and more effective recommended content can be presented. Users can be expected to be enclosed.

  Note that a program for realizing the function of the processing unit in the present invention is recorded on a computer-readable recording medium, and the program recorded on the recording medium is read into a computer system and executed to extract recommended content. May be performed. Here, the “computer system” includes an OS and hardware such as peripheral devices. The “computer system” includes a WWW system having a homepage providing environment (or display environment). The “computer-readable recording medium” refers to a storage device such as a flexible medium, a magneto-optical disk, a portable medium such as a ROM and a CD-ROM, and a hard disk incorporated in a computer system. Further, the “computer-readable recording medium” refers to a volatile memory (RAM) in a computer system that becomes a server or a client when a program is transmitted via a network such as the Internet or a communication line such as a telephone line. In addition, those holding programs for a certain period of time are also included.

  The program may be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by a transmission wave in the transmission medium. Here, the “transmission medium” for transmitting the program refers to a medium having a function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line. The program may be for realizing a part of the functions described above. Furthermore, what can implement | achieve the function mentioned above in combination with the program already recorded on the computer system, what is called a difference file (difference program) may be sufficient.

DESCRIPTION OF SYMBOLS 1 Service provision system 100 User terminal 200 Service provision apparatus 210 Content storage part 220 Service provision part 300 Recommended content extraction apparatus 310 Storage part 311 Usage history storage part 312 User usage history matrix storage part 313 User / centroid similarity matrix storage part 314 Clustering Information storage unit 320 Usage history acquisition unit 330 User usage history matrix generation unit 340 Similarity calculation unit 350 Integrated similarity calculation unit 360 Clustering unit 370 Recommended target user identification information input unit 380 Recommended target service identification information input unit 390 Recommended content extraction unit

Claims (5)

A recommended content extraction device connected to a plurality of service providing devices that transmit content requested from the user terminal among a plurality of defined content to a plurality of user terminals connected via a network,
For each service identification information for identifying the service providing apparatus, user identification information for identifying the user of the user terminal is associated with content identification information for identifying the content transmitted to the user terminal by the service providing apparatus. A usage history storage unit that stores the received usage history information;
A clustering information storage unit that stores, for each user identification information, cluster identification information that identifies the user's cluster determined by clustering based on the usage history information corresponding to the user identification information;
For each of the service identification information, based on a predetermined distance scale, a similarity calculation unit that calculates the similarity of the usage history information between the user and the cluster centroid,
An integrated similarity calculating unit that calculates an integrated similarity that is an average value of the plurality of similarities calculated for each of the service identification information by the similarity calculating unit;
Clustering for each user is performed based on the integrated similarity calculated by the integrated similarity calculating unit to determine a cluster corresponding to the user, and user identification information for identifying the user and corresponding to the user identification information A clustering unit that associates and stores the cluster identification information for identifying the cluster to be stored in the clustering information storage unit;
A recommendation target user identification information input unit for inputting recommendation target user identification information for identifying a user terminal of a recommendation target user who is a target of recommending the content among the plurality of user terminals;
A recommendation target service identification information input unit for inputting recommendation target service identification information for identifying the service provision device that transmits the recommended content to the user terminal of the recommendation target user among the plurality of service provision devices;
The cluster identification information associated with the recommendation object user identification information input to the recommendation object user identification information input unit is read from the clustering information storage unit, and the user identification information associated with the read cluster identification information The content identification stored in the usage history storage unit in association with the read user identification information and the recommendation target service identification information is read out of the user identification information different from the recommendation target user identification information A recommended content extraction unit that determines that the information is content identification information recommended to the recommendation target user;
A recommended content extraction apparatus comprising: The integrated similarity calculation unit multiplies the plurality of similarities calculated for each of the service identification information by the similarity calculation unit by a weighting coefficient predetermined for each of the service identification information, and calculates the weighting coefficient. The recommended content extraction apparatus according to claim 1, wherein the integrated similarity that is an average value of the multiplied similarities is calculated. The use degree information between the users based on any one of a plurality of distance measures including a Euclidean distance and a cosine similarity predetermined for each service identification information. The recommended content extraction device according to claim 1, wherein the similarity is calculated. A user that identifies a user of the user terminal for each service identification information that identifies a plurality of service providing apparatuses that transmit requested content from a plurality of user terminals connected via a network among a plurality of defined contents For each user identification information, a usage history storage unit that stores identification information and usage history information associated with content identification information for identifying the content transmitted to the user terminal by the service providing device, A recommendation content extraction device comprising: a clustering information storage unit that stores cluster identification information that identifies the user's cluster that is determined by clustering based on the usage history information corresponding to the user identification information; A content extraction method,
A step of calculating a similarity of the usage history information between the user and the cluster centroid based on a predetermined distance scale for each service identification information;
An integrated similarity calculating unit calculating an integrated similarity that is an average value of the plurality of similarities calculated for each of the service identification information by the similarity calculating unit;
A clustering unit performs clustering for each user based on the integrated similarity calculated by the integrated similarity calculating unit, determines a cluster corresponding to the user, user identification information for identifying the user, and the user Associating and storing the cluster identification information for identifying the cluster corresponding to the identification information in the clustering information storage unit;
A step of receiving recommendation target user identification information for identifying a user terminal of a recommendation target user who is a target of recommending the content among the plurality of user terminals, to the recommendation target user identification information input unit;
A recommendation target service identification information for identifying the service providing apparatus that transmits content recommended to the user terminal of the recommendation target user among the plurality of service providing apparatuses is input to the recommendation target service identification information input unit; ,
The recommended content extraction unit reads the cluster identification information associated with the recommendation target user identification information input to the recommendation target user identification information input unit from the clustering information storage unit, and associates the cluster identification information with the read cluster identification information Among the received user identification information, the user identification information different from the recommendation target user identification information is read, and the user history information is associated with the read user identification information and the recommendation target service identification information in the usage history storage unit. Determining the stored content identification information as content identification information recommended to the recommendation target user;
A recommended content extraction method comprising: A user that identifies a user of the user terminal for each service identification information that identifies a plurality of service providing apparatuses that transmit requested content from a plurality of user terminals connected via a network among a plurality of defined contents For each user identification information, a usage history storage unit that stores identification information and usage history information associated with content identification information for identifying the content transmitted to the user terminal by the service providing device, A cluster of a recommended content extraction apparatus comprising: a clustering information storage unit that stores cluster identification information that identifies the user's cluster determined by clustering based on the use history information corresponding to the user identification information; In addition,
A step of calculating a similarity of the usage history information between the user and the cluster centroid based on a predetermined distance scale for each service identification information;
An integrated similarity calculating unit calculating an integrated similarity that is an average value of the plurality of similarities calculated for each of the service identification information by the similarity calculating unit;
A clustering unit performs clustering for each user based on the integrated similarity calculated by the integrated similarity calculating unit, determines a cluster corresponding to the user, user identification information for identifying the user, and the user Associating and storing the cluster identification information for identifying the cluster corresponding to the identification information in the clustering information storage unit;
A step of receiving recommendation target user identification information for identifying a user terminal of a recommendation target user who is a target of recommending the content among the plurality of user terminals, to the recommendation target user identification information input unit;
A recommendation target service identification information for identifying the service providing apparatus that transmits content recommended to the user terminal of the recommendation target user among the plurality of service providing apparatuses is input to the recommendation target service identification information input unit; ,
The recommended content extraction unit reads the cluster identification information associated with the recommendation target user identification information input to the recommendation target user identification information input unit from the clustering information storage unit, and associates the cluster identification information with the read cluster identification information Among the received user identification information, the user identification information different from the recommendation target user identification information is read, and the user history information is associated with the read user identification information and the recommendation target service identification information in the usage history storage unit. Determining the stored content identification information as content identification information recommended to the recommendation target user;
Recommended content extraction program to execute

Images